A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Bugfix] Set `VLLM_ALLREDUCE_USE_SYMM_MEM` default to False (#24696)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Sep 11, 2025 at 17:32 UTC
1ec20355f5eaa52b5965bfa01d2348b5ed818750
1 file changed
+1
-1
vllm/envs.py
+1
-1
@@ -1218,7 +1218,7 @@ environment_variables: dict[str, Callable[[], Any]] = {
1218
1219
# Whether to use pytorch symmetric memory for allreduce
1220
"VLLM_ALLREDUCE_USE_SYMM_MEM":
1221
- lambda: bool(int(os.getenv("VLLM_ALLREDUCE_USE_SYMM_MEM", "1"))),
1221
+ lambda: bool(int(os.getenv("VLLM_ALLREDUCE_USE_SYMM_MEM", "0"))),
1222
1223
# Allows vllm to find tuned config under customized folder
1224
"VLLM_TUNED_CONFIG_FOLDER":